It tends to start before eight. Someone pulls a stock position out of one system, a list of open works orders out of another, and last week’s actual output from a machine log, a supervisor’s notebook, or a pile of paper job travellers that came back from the floor on Friday. None of those three sources agree with, so the first hour of the week goes on deciding which one to believe.

By mid-morning there is a spreadsheet. The week is laid out across it, colour-coded, filtered, adjusted for the jobs everyone already knows are late and for the machine that will be down on Wednesday. By lunchtime it has gone out to production, sales and goods-in, and something quietly important has happened: that spreadsheet is now the schedule. Not the plan sitting in the system. The shop floor works to it, customer service quotes from it, despatch checks against it.

It gets rebuilt the following Monday because by Friday it no longer describes anything real. And the person rebuilding it is usually one of the most capable people in the building, which is the part worth sitting with. They are not inefficient. They are absorbing, by hand, a gap the systems leave open.

Almost every manufacturing business ends up here by accident rather than by decision. The machine controllers were bought to run machines. The planning spreadsheet grew out of one planner’s working method and became institutional. Quality records live where the quality manager needed them, which is often not where anyone else can see them. Job travellers are paper because paper survives the shop floor and works when the network doesn’t.

Each of those choices was sensible at the time. What none of them were designed to do was talk to each other. So, demand sits in one place, capacity in another, and material availability in a third, and there is no mechanism that reconciles all three. When a system cannot reconcile them, a human does, and the only tool flexible enough to hold that reconciliation is a spreadsheet.

This is why the usual responses don’t work. More planning discipline doesn’t help, because the discipline is already there. Nor does hiring another planner, which adds a second reconciliation running in parallel with the first. It is a structural problem wearing the costume of a productivity problem.

The cost rarely shows up as a line on a report, which is precisely why it survives so long. It shows up as a set of recurring frustrations that get treated as the nature of the business.

Quality issues surface after dispatch rather than before it, because the inspection record and the production record were never joined at the point where a pattern would have been visible. True job cost is understood weeks after the job closed. Once labour, scrap, and rework have been reconciled by hand, so the pricing decisions made in the meantime were made on an estimate. Capacity gets promised that doesn’t exist, because the person quoting the date is reading a schedule that was accurate the moment it was built and has been decaying ever since.

Then there is expediting. It starts as the exception and becomes the operating model, the daily stand-up that exists to reshuffle priorities, the shortage that is chased rather than predicted, the changeover that happens because a promise needs rescuing. Expediting is expensive in a way that never gets costed, because it consumes the attention of the people who would otherwise be improving things.

There is a statistic that explains the last two years of manufacturing technology better than any of the commentary around it. Make UK’s 2026 research found that 83% of manufacturers are using AI somewhere in the business, in HR, in finance, in administration. In production, the figure is 11%. In quality control, it’s just 6%.

That gap is not caution, and it is not a shortage of ambition on the operations side. It is data. HR, finance, and admin were digitised decades ago, in systems that store structured, timestamped, queryable records. Feed that to a model, and it has something to learn from. The floor, meanwhile, holds its most valuable information in machine logs nobody exports, in paper travellers, in the tacit knowledge of a planner who knows which line runs slow on long jobs.

So the honest sequence is unglamorous. Connected operational data is not a prerequisite you can buy your way past on the route to AI in production. It is the foundation that everything depends on. Without it, everything struggles to deliver meaningful results. With it, things like finite scheduling, predictive maintenance, and quality analytics become far more reachable.

When planning, production and stock can see each other, the Monday morning changes shape. The planner still opens the week, but they are no longer assembling the picture, the picture is there, and their time goes on the decisions only they can make: which of two late jobs to protect, whether to take the changeover now or run it out, which customer to call before they call you.

The risk to the business is more visible upfront. A material shortfall against a confirmed order becomes visible when it is still a purchasing question, rather than three weeks later when it has become a delivery question. A job trending over its estimate is a conversation while it is running, not a post-mortem. Quality patterns attach to the batch, the machine, and the shift, so the same fault stops being discovered independently four times.

And the more advanced work becomes realistic rather than experimental. Finite scheduling, predictive maintenance and quality analytics are all reasonable things to want, and all of them fail on fragmented data for the same reason a new planner would: there is no coherent history to reason from. Build the foundation, and what once felt ambitious becomes both practical and scalable, creating the conditions for continuous improvement across the business.

None of this needs a transformation programme to investigate. It needs one honest audit, and it starts with a single question: how much of your planning still happens outside the system?

Count the spreadsheets that decisions actually depend on. Count the reports that exist because someone couldn’t get the number any other way. Count the hours per week your best operational people spend assembling information rather than acting on it. If that number is larger than you expected, it is worth understanding why before deciding what to do about it.

For anyone who wants a structured way to work through that assessment, our ERP Route Checklist sets out the questions in order.